Walk into a growing number of Australian GP clinics and specialist rooms this year and you may notice something missing: the doctor hunched over a keyboard while you talk. In its place sits a phone or laptop running an artificial intelligence “scribe”, a tool that listens to the consultation and drafts the clinical note automatically. The technology has spread quickly because it promises to hand clinicians back the hours they lose to paperwork. Now a regulatory question is catching up with it, and the answer could reshape how these tools are built and sold in Australia.
As Information Age reported, the crux of the issue is whether AI medical scribes meet the legal definition of a medical device. If they do, they fall under the oversight of the Therapeutic Goods Administration, the national regulator that assesses everything from pacemakers to diagnostic software before it can be used in Australian healthcare. That classification carries real weight, because a product deemed a medical device must be entered on the Australian Register of Therapeutic Goods and meet conformity requirements that most workflow software never has to think about.
Why the definition matters
The line the TGA draws is not academic. Software earns the “medical device” label when it is intended to diagnose, monitor, predict or treat a condition, rather than simply record or administer information. A scribe that faithfully transcribes what was said and formats it into a note sits comfortably on the administrative side of that line. The grey area opens up the moment these tools start doing more than transcribing. Many already summarise, reorganise and prioritise clinical information, and some vendors are adding features that suggest diagnoses, flag possible drug interactions or propose billing codes. Each step toward clinical judgement nudges the product closer to the regulated category.
That ambiguity is uncomfortable for everyone involved. Vendors building scribes have raced to market on the understanding they are selling productivity software, and a reclassification could force expensive compliance work, fresh documentation and, in some cases, a rethink of the product itself. Clinics that have embedded the tools into daily practice could find themselves relying on something that, on a strict reading, should have been assessed as a device before it ever reached the consulting room.
Two ways to read the risk
There are genuinely competing views on how worried the sector should be. One camp argues that regulation is overdue and welcome. Clinical notes are not trivial documents. They follow a patient between providers, inform decisions years later and carry medico-legal consequences if they are wrong. An AI system that quietly invents a detail, drops a medication or misattributes a symptom is introducing clinical risk, even if it never claims to diagnose anything. On this reading, bringing scribes under a clear regulatory framework protects patients and gives careful vendors a standard to point to, rather than leaving the whole category to operate on trust.
The other camp warns that heavy-handed classification could stall a technology that is demonstrably easing one of healthcare’s most stubborn problems. Clinician burnout driven by administrative load is well documented, and scribes are among the few AI tools in medicine to show an immediate, tangible benefit. If the compliance burden becomes too steep, smaller Australian developers may be squeezed out, leaving the field to large overseas players with the resources to absorb the cost. The fear is not regulation itself but regulation that is slow, uncertain or applied unevenly, leaving developers guessing about where the line sits.
Professional bodies have already tried to fill the gap while the formal position settles. The Royal Australian College of General Practitioners and the Australian Digital Health Agency have both issued guidance urging clinicians to treat scribe output as a draft rather than a finished record, to check every note before signing it, and to be transparent with patients about recording and consent. That guidance is useful, but it is not the same as a clear regulatory status, and it leaves individual doctors carrying much of the responsibility for tools they did not build.
What it means for Australia
For Australia specifically, this is a test of whether the country can regulate applied AI without smothering it. Healthcare is one of the most promising and most sensitive places to deploy the technology, and the decisions made here will echo across aged care, allied health and hospital systems that are watching closely. The federal government has spent the past year signalling a more active stance on AI safety, including proposals around high-risk uses, and medical scribes are a near-perfect example of a tool that feels low-risk day to day but touches genuinely high-stakes records.
There is also a sovereignty dimension that matters to local players. Australian privacy law, including the handling of sensitive health information under the Privacy Act, already imposes obligations on how consultation audio and transcripts are stored and processed. A clear TGA framework, layered on top of those privacy rules, could actually advantage well-run Australian vendors who can demonstrate compliance and keep data onshore, provided the pathway is defined quickly enough for them to plan around it. Left vague, the same uncertainty becomes a competitive disadvantage against international products that arrive with deeper compliance teams.
The patient’s view should not get lost in the regulatory detail either. Most people sitting in a consultation have no idea whether an AI tool is listening, how long the recording is kept, or who can access it. Consent and transparency are not just ethical niceties here, they are likely to become part of whatever framework emerges, and clinics that have not thought carefully about how they explain the technology to patients may find themselves behind.
What happens next
The immediate question is whether the TGA offers clearer, scribe-specific guidance rather than leaving developers to interpret general software-as-a-medical-device rules. A definitive signal, one way or the other, would let vendors plan and let clinics keep using the tools with confidence. Until then, the sensible posture for Australian practices is the one the professional bodies already recommend: treat the AI as an assistant whose work must be checked, be upfront with patients, and keep an eye on where a product sits on the spectrum between transcription and clinical judgement. The technology is not going away. The coming months will decide the rules it has to play by.
Sources: Information Age | ACS.
